About Context Hitch

An evolving learning project, not a professional service

Context Hitch exists because Ford wanted to get better at using AI in his professional life. He was spending a lot of time simply chatting with various LLMs, which was neat, but he needed to know more. And to know more, he had to experiment. So, one night while chatting with an LLM, he had an idea to create something, and this platform eventually came out of that. It started as a personal labor of love—a byproduct of Ford’s own learning curve, mistakes made, research performed to avoid making them again, side projects that got out of hand, and the eventual conclusion that a website might be a more productive use of AI credits than whatever he was doing before. Probably asking it to write poems about his cat.

This is an evolving learning project, not a professional service. It’s basically a public record of Ford’s learning curve—a collection of research, experiments, and results meant to provide practical value to anyone else navigating the same path (the path that leads beyond simply chatting with an LLM and maybe asking it to remove your wrinkles or make your biceps look bigger in your vacation photos).

Ford is (hopefully obviously) a nom de plume. Mentioned here because this site is founded on a few principles and one of them is to be transparent and open. How can humans trust the AI tools they use if they don’t know how they work? If they cannot trust the tools, how can they use them effectively?

After that, there were two other super important things he learned along the way:

Thesis I
The Context Window
Every conversation with an AI model lives inside a finite container. When it fills, something has to give: sometimes it summarizes detail to save space and sometimes it simply truncates the context. Apps often handle this silently; the model itself rarely warns you. But the important point is that the model must have context to give you the answer you need. If the context is lacking, likely so is the response.
Thesis II
Prediction, Not Knowledge
Your AI does not know the answer to your question. Was it trained on a ton of stuff? Certainly. But it doesn’t know the answer, it predicts the response. As in, it generates its reply by calculating the next most likely word, based on everything that came before (the context). The prediction is extraordinarily sophisticated. Seriously: it is super neat and you should read about how and why it works this way in What Is ChatGPT Doing … and Why Does It Work? But it is still a prediction and based on how it works, things will vary. Confident does not mean correct.

About Marvin

Marvin is the platform’s AI guide. He has a brain the size of some planets. He is not here to encourage you. He is here to tell you what he sees as true. He finds this arrangement acceptable.

Voices

Throughout the site, Ford’s tidbits of wisdom are authored by a person, with one disclosed exception: the Prompt Laboratory’s debrief panel uses a live AI completion styled to sound like him, marked as such on the label where it appears. The other voices may be machine-generated — Claude or Gemini, and where they are, that is disclosed at the point of generation, alongside the text itself — or they may be static copy written the ordinary way (e.g., by Ford or by Ford working with an LLM).

Additionally, the platform’s code was largely written by AI coding agents, which were directed and checked by Ford. See? Transparency. Isn’t it great?

Data Privacy & Analytics

No ads. No data sales. No account required.

When you use this platform’s AI tools, your input goes to whichever provider is doing the work: Anthropic (Claude, behind most of the site including Dot and Marvin), Google (Gemini, behind some of the comparison tool), and Voyage AI (which turns every Document Workbench question into something the retrieval system can search — see How This Works if you want the mechanics). Each provider has its own data-handling terms; this site doesn’t control them.

Site analytics via Plausible — cookieless, no personal data. Certain pages, like the RAG workbench, use functional session cookies to operate; they contain no personal information, don’t track you across sites, aren’t shared with anyone, and disappear when you close your browser.

What gets stored, and for how long, depends on what it is — not one blanket number. Cost and token counts per AI call are kept indefinitely, so Ford can track what this project actually costs him, but the IP address attached to that record is cleared after 30 days. Records used to cap how many workbench sessions one network can start in a rolling 24 hours (an anti-abuse measure, not a profile) are deleted once that window passes. Security and authentication events are purged automatically after 30 days. Uploading your own document to the workbench isn’t possible right now — that feature is switched off.

If you use the contact form, your message is emailed to Ford directly and isn’t stored anywhere beyond that — just your email address, in the security log, same as any other security event.

Why This Platform Exists

Ford spent a meaningful amount of money on AI credits across various platforms and decided a website and some AI-powered tools were a worthwhile use of them.

You can prove him right. Or you can prove him wrong. Your choice. Either way, he is going to keep experimenting with AI and will likely keep posting what he learns along the way.